Star-P

Star-P is targeted to scientists, engineers, and researchers, and provides the ability to bridge their use of desktop coding tools directly to parallel, multi-CPU environments. Specifically, end users can develop computational prototypes using MATLAB or Python on their desktop machines, and execute those models in high performance parallel processing environments without first requiring that the models be reprogrammed in C, Fortran, using an MPI.

To use Star-P, the researcher continues to use MATLAB or Python; a Star-P Client is loaded on the researcher's machine and transparently intercepts function calls and commands, routing execution to the parallel environment as needed. Instructions can be emdedded into the code enabling the automatic execution of that code on the parallel server. Both fine and coarse grain (also called parallel data execution, where the individual processors in the computation are able to communicate with one another; and parallel task execution, where the processors needn't communicate with one another but rather perform tasks simultaneously) processing is available. Fine grained execution is available via the addition of a "*p" construct on the designated variables (I.E., computations on variables designated with a *p are automatically sent to the parallel server for execution, and any related variables are also automatically processed on the parallel server through propagation); and coarse-grained parallelism provides for the automated processing of parallel tasks (such as unrolling serial FOR loops) independent of the number of processors the program has access to (i.e., Star-P automatically distributes the data, executes the computations, gathers the results, and returns them to the calling application).

For Python in particular, the tool operates somewhat as a drop-in replacement for NumPy; instead of importing the NumPy package, the user would import the starp package, connect to the parallel server (using an included defaultConnect method), and then reference NumPy methods exactly as they would normally, only through the starp class (i.e., starp.numpy.random.rand(5, 5)).

Other features of the platform include an administration interface, allowing for the creation and modification of user and session profiles (connection parameters, access rights, user privileges, assigned CPU sets, system monitoring, etc.); and a profiling tool for the optimization of designed code.

New features of Star-P include support for Platform Computing's LSF (a management/scheduling tool for parallel application execution), and SGI Altix ICE blade server support.

Star-P is available now. The base price for an 8 socket system is $15,960 for a yearly subscription.

Visit the Interactive Supercomputing Web site for further information.